Add transient-network retry with exponential backoff

- bls_client/retry.py: with_retries() retries timeouts/connection drops and
  429/5xx with exponential backoff; honors Retry-After; never retries 4xx or
  BLSQuotaError (those won't fix themselves)
- BLSClient(retries=2, backoff=0.5) wraps the data POST and discovery GETs;
  qcew CSV fetch wrapped too. retries=0 disables.
- tests/test_retry.py: 6 offline tests (injectable sleep, no real delays)
- README: retry behavior documented; drop the now-resolved limitation

Offline suite 125 passing; live smoke green.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

bls_client/retry.py

@@ -0,0 +1,58 @@
+"""Retry transient network failures with exponential backoff.
+
+Shared by the JSON client (POST) and the QCEW CSV client (GET). Retries only
+genuinely transient conditions — connection drops, timeouts, and 429/5xx — never
+4xx client errors or a BLS quota rejection (those won't fix themselves).
+"""
+import time
+
+import requests
+
+# HTTP statuses worth retrying (429 = throttled, 5xx = server-side transient).
+RETRY_STATUS = {429, 500, 502, 503, 504}
+
+# Network-level exceptions worth retrying.
+RETRY_EXCEPTIONS = (
+    requests.exceptions.ConnectionError,
+    requests.exceptions.Timeout,
+    requests.exceptions.ChunkedEncodingError,
+)
+
+
+def _retry_after(resp) -> float | None:
+    """Honor a numeric Retry-After header if the server sent one."""
+    ra = getattr(resp, "headers", {}).get("Retry-After") if resp is not None else None
+    if ra and str(ra).isdigit():
+        return float(ra)
+    return None
+
+
+def with_retries(send, *, retries=2, backoff=0.5, backoff_max=8.0, sleep=time.sleep):
+    """
+    Call ``send()`` (which returns a requests.Response), retrying transient
+    failures. Backoff is ``backoff * 2**attempt`` seconds, capped at ``backoff_max``.
+
+    Args:
+        send:        zero-arg callable issuing the request and returning a Response.
+        retries:     max retries after the first attempt (so retries=2 -> 3 tries).
+        backoff:     base backoff in seconds.
+        backoff_max: cap on a single backoff interval.
+        sleep:       sleep function (injectable for tests).
+    """
+    attempt = 0
+    while True:
+        try:
+            resp = send()
+        except RETRY_EXCEPTIONS:
+            if attempt >= retries:
+                raise
+            sleep(min(backoff * (2 ** attempt), backoff_max))
+            attempt += 1
+            continue
+
+        if getattr(resp, "status_code", None) in RETRY_STATUS and attempt < retries:
+            sleep(_retry_after(resp) or min(backoff * (2 ** attempt), backoff_max))
+            attempt += 1
+            continue
+
+        return resp